Leader Election Algorithms for Multi-channel

Wireless Networks

Tarun Bansal, Neeraj Mittal, and S. Venkatesan

Department of Computer Science

The University of Texas at Dallas
Richardson, TX 75080, USA
tarun@student.utdallas.edu, {neerajm,venky}@utdallas.edu

Abstract. We study the leader election problem in single-hop multi-

channel wireless networks with single-antenna radio nodes. The objec-
tive is to elect leaders for all channels on which one or more nodes in
the network can operate. We assume that nodes do not have collision de-
tection capability. In this paper, we propose three algorithms for leader
election: one deterministic and two randomized. The deterministic algo-
rithm executes for at most 2N + M log M  time slots and guarantees
leader election for all the channels, where M is the size of the univer-
sal channel set available to the nodes and N is the size of the label
space used to assign unique labels to the nodes. The randomized al-
gorithms guarantee that a leader is elected for all the channels with
probability at least 1 − 1/f within O(M log 2 (nmax ) + M log2 (M f )) and
O(M log(nmax ) log(M f )) time slots, respectively, where nmax is the max-
imum number of nodes operating on any channel in the network. To the
best of our knowledge, this is the first work on leader election in single-
antenna multi-channel radio networks.

Keywords: wireless networks, cognitive radios, multiple channels, leader

election.

1 Introduction

Currently, the wireless channels are allocated using the fixed spectrum alloca-
tion policy in which the spectrum is assigned by a central authority to various
services. However some frequencies are much more in demand than other owing
to their lower energy requirements, low error rate and higher transmission range.
As a result, a large portion of the spectrum is sporadically used whereas in some
portions the usage is a lot more concentrated [1]. In order to allocate channels
efficiently so as to avoid spectrum holes, the use of cognitive radios has been
recommended [1,2].
Cognitive radios (CRs) are able to change their transmission and reception
parameters to communicate efficiently avoiding interference with other users by
sensing the spectrum. This adjustment can be done dynamically and can in-
clude changes in communication frequency, encoding, link layer parameters etc.

Y. Li et al. (Eds.): WASA 2008, LNCS 5258, pp. 310–321, 2008.


c Springer-Verlag Berlin Heidelberg 2008
 Leader Election Algorithms for Multi-channel Wireless Networks 311

Even though cognitive radios are equipped with only a single antenna, still this
dynamic adjustment allows them to operate on multiple channels.
While availability of multiple channels allow better network throughput po-
tentially, but it also makes the setting up of the network much more difficult.
This is because different nodes may have different capabilities. For example, a
node R1 may be able to communicate on channel C5 , but a neighboring node R2
may not be able to communicate on that channel because of different hardware
capability or spatial variance of the channel availability set [1]. This makes it a
lot more difficult to solve traditional problems for multi-channel radio networks.
Leader election is one such fundamental problem. It involves selecting a distin-
guished node in the network. In wireless networks, leader election has been used
to solve many other problems like routing [3], key distribution [4], coordination
in sensor networks [5], neighbor discovery [6,7] and so on.
For multi-channel networks, a leader can be elected in two different ways:
(i) one leader for all the channels, or (ii) one leader on each channel with possibly
different leaders for different channels. Having one leader for all the channels
is not feasible if no node in the network has the capability to communicate
on all the channels. Therefore, to ensure that every node in the network can
communicate with a leader of a channel in a single step, we have to ensure that
there is a leader for every channel on which one or more nodes in the network
can operate.
In this paper, we propose several leader election algorithms (one determin-
istic and two randomized) for single-hop multi-channel networks (like cognitive
radio networks). The deterministic algorithm guarantees that a leader is elected
for all channels using at most 2N + M log M  time slots, where M is the size
of the universal channel set available to the nodes and N is the size of the
label space used to assign unique labels to the nodes. Note that the time com-
plexity of our deterministic algorithm is much lower than that of the naı̈ve
algorithm, which uses M N time slots. The randomized algorithms guarantee
that a leader is elected for all the channels with probability at least 1 − f1 within
O(M log2 (nmax )+M log2 (M f )) and O(M log(nmax ) log(M f )) time slots, respec-
tively, where nmax is the maximum number of nodes operating on any channel in
the network. Both randomized algorithms work without assuming the knowledge
of nmax or N . As it can be observed, the time complexity of both randomized al-
gorithms is asymptotically much lower than that of the deterministic algorithm
when N  M log2 (nmax ) + M log2 (M f ).
The rest of the paper is organized as follows. In section 2, we discuss the
previous work conducted on leader election in single channel networks. We
describe our system model in section 3. In section 4 we describe a determin-
istic algorithm for leader election. In section 5, we describe two randomized
algorithms for leader election. Finally we conclude the paper in section 6. Due
to space constraints, some of the proofs have been omitted and can be found
elsewhere [8].
312 T. Bansal, N. Mittal, and S. Venkatesan

2 Related Work
Depending on radio transceiver’s capability to detect collision, two radio models
have been used in literature [9] for developing wireless algorithms viz. “Strong
Radio model” and “Weak Radio Model”. In the weak radio model, nodes cannot
distinguish between “no transmission” and “multiple transmissions”. This is also
known as Radio Network with no Collision Detection (no-CD RN). Moreover,
the transmitting nodes are not capable of monitoring the state of the channel
simultaneously while transmitting. In strong radio model, nodes have collision
detection capability which allows them to differentiate between “no transmis-
sion” and “multiple transmissions”. Strong radio model also allows transmitting
nodes to detect collision by simultaneously allowing them to monitor the channel
state. As pointed out by the authors [9], only the weak radio model is consistent
with the IEEE 802.11 standards and current WLAN standards.
For deterministic leader election algorithms for single channel weak radio net-
works, Jurdzinski et al. [10] have proved that the lower bound on time require-
ments for such algorithms is N − 1 time slots. Nakano et al. [11] have given one
such time optimal and energy efficient deterministic algorithm which terminates
in optimal number of time slots.
A trivial way to design deterministic algorithm for leader election for multiple
channels could be to run M instances of Nakano’s [11] algorithm on each channel.
However, in that case the time complexity of the protocol would become O(N M ).
In this paper, we present a deterministic algorithm for leader election which
makes use of FDMA to achieve a much lower time complexity of O(N +M log M ).
Different randomized algorithms for leader election have been proposed in the
literature for different radio models. The leader election algorithms proposed in
[10], [12], [13], [14] are based on strong radio model. They either assume that the
nodes have collision detection capability or that the nodes can monitor the state
of the channel at the same time while transmitting. In this paper, we consider
only those algorithms that assume weak radio model.
Metcalfe et al. [15] have proposed algorithms for leader election which assumes
the knowledge of the number of nodes present in the network. However for multi-
channel networks, it is possible that the number of nodes present on different
channels differ vastly. This may be due to different hardware capabilities of the
devices or because of the spatial variance of the channel availability set [1]. For
example, a higher percentage of nodes may be able to communicate on channels
which have low energy requirements, low error rate etc. as compared to channels
which do not exhibit these properties. This makes it very difficult to predict the
number of nodes on each channel as required by their algorithm.
Hayashi et al. [16] have given a randomized algorithm for leader election for the
weak radio model with unknown number of nodes present on the channel, which
terminates in O(log2 n) broadcast rounds with probability exceeding 1 − 1/n
where n is the actual number of nodes present on the channel. Their algorithm
proceeds in multiple rounds and each round is further divided into time slots. The
number of time slots in a round increase with the round number. The ith round
of execution is divided into i time slots. At the beginning of each round, nodes
 Leader Election Algorithms for Multi-channel Wireless Networks 313

set the probability of transmission as 1/2 and with each time slot in that round,
the probability of transmission is reduced by half. The algorithm terminates as
soon as a time slot occurs in which exactly a single node transmits.
For the networks where the number of radio nodes present is not known,
Nakano et al. [11] have also proposed three different algorithms for leader elec-
tion. The first algorithm proposed by them terminates with probability exceeding
1−1/f in O((log2 n)+(log2 f )) time slots. One of our randomized leader election
algorithm is derived from this algorithm.

3 System Model
We consider leader election problem for one hop radio network with weak radio
model. We term the set of all the channels as universal channel set, Auniv . The
nodes can communicate on only a subset of the channels of the global channel
set. So two nodes Ri and Rj would be termed as neighbors only if there exists
some channel Ck such that both Ri and Rj can communicate on Ck . We assume
that the nodes have been assigned unique labels from the set: {1, 2, 3, ..., N }.
In this paper we give a deterministic and a randomized algorithm for leader
election in multi channel networks. Unless otherwise stated, we assume that all
nodes are aware of the global channel set and the size of the label space. We use
the following notation to describe our algorithms:
- M = Number of channels in the universal channel set
- Auniv = Universal channel set
- N = Size of the label space assigned to the nodes
- Ci = Channel i from the universal channel set Auniv
- Ri = Radio node with label i
- Ai = Set of channels available to node Ri , if Ri is present in the network
- nmax = Maximum number of nodes present on any channel
Observe that N acts as an upper bound on nmax . However, in practice, nmax
may be much smaller than N . Also, observe that there may be one or more
channels on which no node can operate. Clearly, a leader cannot be elected for
such channels. Therefore we call a channel relevant if there are one or more
nodes in the network that can operate on that channel.

4 Deterministic Algorithm for Leader Election

In this section, we propose a deterministic leader election algorithm which runs
in multiple phases. With each phase, on each channel, we reduce the label space
of the competing nodes by a factor of 2. Initially, the label space of the nodes
competing to become leader is N for each channel. When log N phases have
completed execution, the label space is reduced to a unit size for each channel.
Also, we ensure that for every channel, there is exactly one node with that label
in the entire network. This unique node is then elected as the leader of the
corresponding channel.
314 T. Bansal, N. Mittal, and S. Venkatesan

Phase 1 ··· Phase i ···

 
N
Round 1 ··· Round j ··· Round 2i M

M time slots

Fig. 1. Phases and rounds

Figure 1 gives the high level overview of the working of the algorithm. The
execution of the algorithm is divided into log N phases and each phase is further
divided into k rounds where k depends on the label space size at the beginning
of the round. Further, each round takes M time slots to complete.

Working within a phase: Let us assume that at the beginning of the phase, the
size of the label space of the competing nodes is X. So we divide this label space
into groups of size 2M . There are  2M
X
 such groups. We then further divide the
execution of the phase into  2M  rounds.
X

In each round, the algorithm handles one block of label space of size 2M . In
each round, we reduce the size of label space of competing nodes from 2M to
M . So if at the start of the round, all the label space was occupied by nodes
such that there were up to 2M nodes competing to become leader, we ensure
that when the execution of the round terminates, out of the 2M nodes at most
M nodes are still qualified for being elected as leaders. Thereafter, only these
M nodes, get the chance to participate in the leader election process in the next
phase. The remaining M nodes are no longer eligible for leader election on that
channel.
However, if at the beginning of the phase, number of competing nodes is less
than 2M , then the phase would complete in a single round. Note that in this
case as well, the algorithm would be able to reduce the label space size by a
factor of 2.

Working of a round: At the beginning of each round, we group the 2M nodes

into pairs of two and then these M node-pairs can resolve priority between
themselves on all the channels in M time slots. Only the node with the higher
prioirty, gets the chance to compete for leader election on that channel in the
next round.
Table 1 shows the working of the protocol within each round. We make use of
FDMA so as to reduce the number of time slots required to complete the exe-
cution of one round. This helps us reduce the time complexity of the algorithm.
Also, note that the schedule is designed in such a way that all the M node-pairs
get the chance to decide priority between themselves on every channel. The table
is shown for the case when the range of labels of the nodes in the group is from 1
 Leader Election Algorithms for Multi-channel Wireless Networks 315

Table 1. Shows the execution of the algorithm within a round when label space of
participating nodes is {1, 2, ..., 2M }. The addition is done modulo M . In case there
are some labels with no nodes, then the time slots corresponding to those labels remain
blank.

Channel
C C2 ... Cj ... CM −1 CM
Time Slot 1
1 1, 2 3, 4 2j − 1, 2j 2M − 2M − 1, 2M
3, 2M − 2
2 2M − 1, 2M 1, 2 2j − 3, 2j − 2 2M − 5, 2M − 3,
2M − 4 2M − 2

i 2M − 2i + 3, 2M − 2i + 5, 2M −2i+2j +1, 2M − 2i − 1, 2M − 2i + 1,
2M − 2i + 4 2M − 2i + 6 2M − 2i+ 2j + 2 2M − 2i 2M − 2i + 2

M −1 5, 6 7, 8 2j + 3, 2j + 4 1, 2 3, 4
M 3, 4 5, 6 2j + 1, 2j + 2 2M − 1, 2M 1, 2

to 2M . However, if the participating nodes have labels in a different range (say

x + 1 to x + 2M ), then in order to figure out the transmission schedule for them,
we first need to change their range temporarily from 1 to 2M . This can be done
trivially by subtracting a suitable number (x in this case) from all the labels.
So from the table we can see that for example in slot 1, nodes belonging to
pair 1, viz. nodes 1 and 2, decide priority between themselves on channel C1 .
Also at the same time the nodes belonging to pair 2, viz. nodes 3 and 4, decide
priority on channel C2 .
However, if at the beginning of the round, the number of nodes competing to
become leader on the channel is less than 2M , the time taken to run the round
is still M time slots.
To decide priority between two nodes (Ri and Rj ) in a single time slot on
some channel (Ck ), we use the short routine given in Algorithm 1. The node
which has higher priority moves to the next phase on channel Ck . Note that if
there is no node with either of the labels, then the time slot goes empty and
none of them is able to move to the next phase.
Also the nodes which move to next phase reduce their label by a factor of 2
and if it comes out to be fractional it is rounded up. This helps us to reduce
the size of the label space by 2 with each phase. Note that the way node-pairs
are formed, it is guaranteed that on any channel, two nodes with same label will
never compete to become the leader. Since the size of the label space would be
smaller in next phase, the number of rounds in the next phase would also reduce
by a factor of 2. (Since number of rounds in a phase is proportional to the size
of the label space at the beginning of the phase). However this trend continues
only as longs as the number of competing nodes is more than 2M . After that,
each phase contains exactly one round. We will now prove the correctness of the
algorithm.
316 T. Bansal, N. Mittal, and S. Venkatesan

Algorithm 1. decide priority

if Ri .exists and Ck ∈ Ai then
Ri sends beacon on Ck , Ri knows it has higher priority between Ri and Rj .
if Rj .exists and Ck ∈ Aj then
Rj receives beacon from Ri . Rj knows Ri has higher priority.
end if
else if (¬Ri .exists or Ck ∈
/ Ai ) and (Rj .exists and Ck ∈ Aj ) then
Rj does not receive any beacon, then Rj has higher priority between Ri and Rj
else if (¬Ri .exist or Ck ∈
/ Ai ) and (¬Rj .exists or Ck ∈
/ Aj ) then
None of the nodes can communicate on Ck . So none of them move to the next
phase.
end if

Lemma 1. On any channel, no two nodes which are competing to become leader
on that channel can have the same label.

Lemma 2. Leader is elected for all the channels wherever possible.

Lemma 3. For each channel, at most one leader is elected.

Lemma 4. The number of rounds in phase i is  2i ×M

N
.

Theorem 1. The deterministic algorithm takes at most 2N + M log M  time

slots for completion.

Proof. Phase i has  2i ×M

N
 rounds and each round takes M time slots for execu-
tion. Therefore the time required for completing Phase i would be  2i ×M
N
 × M.
Hence, total time required would be:

X 
 
N
×M (1)
i=1
2 ×M
i

where X is the number of phases in the execution. For values of N of the form
M × 2t (where t ∈ N), X = t + log M . The first t phases reduce the label
space size from N to M and the remaining log M  phases reduce the label
space size from M to 1. It can be verified that this summation evaluates to
N − M + M log M . For general values of N with N > M which are not of this
form, we pick the smallest integer N  of the form M × 2t (where t ∈ N) that is at
least N . This new integer N  is then considered as the new N . The time required
for completion would be N  − M + M log M  time slots. Since N  ≤ 2N , the
execution of the deterministic protocol takes at most 2N − M + M log M  time
slots.
Finally, for values of N and M such that N ≤ M , it can be shown that
summation simplifies to M log N , which is at most M log M . 

 Leader Election Algorithms for Multi-channel Wireless Networks 317

Table 2. Description of the execution of the protocol over the multiple phases. Column
2 shows the range of original labels possible for the nodes having the corresponding
labels in Column 1 during the particular phase. (a) Phase 1 execution: Six labels (or 3
rows of the table) are processed in 1st round of execution. Only 4 labels are processed in
the 2nd round. (b) Phase 2 execution, (c) Phase 3 execution, and (d) Phase 4 execution.

Label Original Label C1 C2 C3

1&2 1-2 1
3&4 3-4 3 3 Label Original Label C1 C2 C3
5&6 5-6 6 1&2 1-4 1 3
7&8 7-8 7 3&4 5-8 6 7
9 & 10 9 - 10 9 5 9 - 10 9
(a) (b)

Label Original Label C1 C2 C3

1&2 1-8 1 3 7 Label Original Label C1 C2 C3
3 9 - 10 9 1&2 1 - 10 1 3 7
(c) (d)

Example: We now show the working of the deterministic algorithm through an

example. Let us assume that there are 3 channels and 5 nodes in the network.
The initial size of label space is 10. The channel availability set of the nodes is:
A1 = {C1 }, A3 = {C1 , C2 }, A6 = {C2 }, A7 = {C3 }, A9 = {C3 }.
During the first phase, we group the label space into two groups of size 6 and
4 respectively. The transmissions during phase 1 have been given in Table 2(a).
This phase takes 2 rounds to complete. We have shown the execution during both
the rounds of Phase 1. Since during the phase 1, labels of the nodes are same
as their original labels, therefore the values in column 1 and 2 are same. The
values in the cells are the original labels of the node which had higher priority
during the execution of decide priority() on that channel. For example, R1 can
communicate on channel 1, therefore it has higher priority over R2 (which does
not exist in this case) on this channel.
Similarly, when the second phase terminates, the label space size would reduce
to 3 (see Table 2(b)). This time the phase execution would complete in a single
round. Since the node labels have now reduced by a factor of 2, therefore a node
which has label 2 would actually have its original label as 3 or 4. Also note that
even though both the nodes R1 and R3 were able to move to Phase 2 on channel
1, still only R1 would move to Phase 3 as given by decide priority().
In the 3rd phase, the label space size would reduce to 2. This time also the
phase execution would complete in a single round. (see Table 2(c)). On termi-
nation of the 4th phase, the label space size would reduce to 1. The execution
proceeds in a single round as shown in Table 2(d).
Total rounds required for the execution (over all the phases) = 2 + 1 + 1 +
1 = 5. Each round takes 3 time slots to complete. Hence total execution time is
318 T. Bansal, N. Mittal, and S. Venkatesan

15 time slots. Note that if we had elected leaders the trivial way, the total time
slots required would have been N M = 10 ∗ 3 = 30 time slots.

5 Randomized Algorithms for Leader Election

In this section, we propose two different randomized algorithms for leader elec-
tion. Both the algorithms work even when an upper bound on the number of
nodes in the network is not known. We refer to the first algorithm as Rand-Elect
and to the second as Fast-Rand-Elect.

5.1 Algorithm Rand-Elect

This algorithm for multi-channel leader election uses the algorithm proposed by
Nakano and Olariu in [11] as a subroutine. Readers are referred to [11] for a
detailed description. Here, we present a short overview of their algorithm. Their
algorithm proceeds in multiple phases. At the beginning of each phase, nodes
start with probability of transmission as 1/2, and with each slot within the
phase, the probability of transmission is reduced by a factor of 2. The number
of time slots in a phase are gradually incremented by one. Therefore, phase i
consists of i time slots. And, in time slot j of phase i, nodes transmit with
probability 1/2j (and listen with probability 1 − 1/2j ). At the beginning of
next phase, the probability of transmission is again reset to 1/2. This pattern is
continued until some time slot occurs in which exactly one node transmits and all
other nodes listen. As soon as that happens, the unique node that transmitted
in this time slot is elected as the leader of the network. Nakano and Olariu’s
algorithm gurantees that a leader is elected with probability at least 1− f1 within
O(log2 n + log2 f ) time slots, where n is the actual number of nodes present in
the network [11].
To solve the leader election problem for multiple channels, we propose to run
M instances of their algorithm on each of the M channels concurrently. The
multi-channel leader election algorithm executes in rounds. Each round consists
of M time slots. In the ith time slot of every round, we simulate a single step of
the ith instance of the algorithm, which is running on channel Ci . Observe that
only nodes with channel Ci in their channel availability set participate in the ith
instance of the algorithm.

Theorem 2. Let nmax denote the maximum number of nodes that can be present
on any channel in the network. Then Rand-Elect elects leader for all relevant
channels within O(M log2 (nmax ) + M log2 (M f )) time slots with probability at
least 1 − f1 .

Proof. Since (1 − Mf ) ≥ 1 − f1 , therefore in order to elect leaders for all

1 M

relevant channels with probability exceeding 1 − f1 , it is sufficient to ensure that

the probability of leader election on individual channels exceeds 1 − Mf 1
. Now,
th
the i instance of Nakano and Olariu’s algorithm guarantees leader election on
 Leader Election Algorithms for Multi-channel Wireless Networks 319

channel Ci with probability at least 1− Mf 1

within O(log2 (ni )+log2 (M f )) steps,
where ni is the number of nodes that can operate on channel Ci . Since one round
simulates one step of each instance of Nakano and Olariu’s algorithm, Rand-Elect
guarantees leader election for all relevant channels with probability at least 1− f1
within O(M log2 (nmax ) + M log2 (M f )) time slots. Hence the result. 


5.2 Algorithm Fast-Rand-Elect

We now present another randomized leader election algorithm, which we call as
Fast-Rand-Elect, that is faster than Rand-Elect in many cases. Before we describe
the working of the new algorithm, we first describe a leader election algorithm
(Alg-Known-Size) for single channel networks when the number of nodes present
is known beforehand. The pseudo code for Alg-Known-Size is given in Algorithm
2 where S is the number of nodes present in the network and T is the maximum
number of time slots for which the algorithm can run. In the pseudo-code, t
denotes the sequence number of the current time slot.

Algorithm 2. Alg-Known-Size(S, T)
t←1
repeat
transmit with probability S1
t ← t+1
until exactly one node transmits or t > T

It can be shown that Alg-Known-Size guarantees leader election with proba-

bility at least 1 − eT1/4 [15], [16], [11]. Now, suppose we wish to elect a leader for
all relevant channels in a multi-channel network with probability at least 1 − f1 .
Clearly, it is sufficient to ensure that a leader is elected for each relevant channel
with probability at least 1 − Mf 1
because (1 − Mf ) ≥ 1 − f1 .
1 M

To elect leaders for all relevant channels in a multi-channel network when

the network size (or even an upper bound on network size) is not known, we
run multiple instances of Alg-Known-Size on each of the channels one-by-one
with geometrically increasing values of network size. The pseudo-code of the
algorithm Fast-Rand-Elect is given in Algorithm 3. Specifically, the execution of
Fast-Rand-Elect is divided into multiple rounds. In round x, we run an instance
of Alg-Known-Size on each of the channels one-by-one using an estimate of 2x
for network size. Again, as before, only those nodes in the network that can
operate on channel Ci participate in the instance of Alg-Known-Size running on
channel Ci . In the pseudo-code, i denotes the channel number and x denotes the
current round number.
We now prove the correctness of the algorithm.
Lemma 5. Let ni denote the number of nodes in the network that can operate
on channel Ci with ni ≥ 1. Then, by the end of log(ni ) rounds, Fast-Rand-Elect
elects a leader on channel Ci with probability exceeding 1 − Mf1
.
320 T. Bansal, N. Mittal, and S. Venkatesan

Algorithm 3. Fast-Rand-Elect for multi-channel networks

x←1
loop
for i = 1 to M do
run an instance of Alg-Known-Size( 2x , 8 ln(M f )) on channel Ci
end for
x ←x+1
end loop

Theorem 3. Let nmax denote the maximum number of nodes present on any
channel in the network. Then Fast-Rand-Elect elects a leader on all relevant
channels with probability exceeding 1 − f1 within O(M log(nmax ) log(M f )) time
slots.

Proof. From Lemma 5, Fast-Rand-Elect elects a leader on a relevant channel with

probability at least 1 − Mf
1
within O(log(nmax )) rounds. Each round consists of
O(M log(M f )) time slots. Therefore Fast-Rand-Elect elects a leader on a relevant
channel with probability at least 1 − Mf 1
within O(M log(nmax ) log(M f )) time
slots. This, in turn, implies that Fast-Rand-Elect elects a leader on all relevant
channels with probability at least 1 − f1 within O(M log(nmax ) log(M f )) time
slots. 


Observe that Fast-Rand-Elect has better asymptotic time complexity than Rand-
Elect if log(nmax ) is either o(log(M f )) or ω(log(M f )). In other words, log(nmax )
∈ Θ(log(M f )). In case log(nmax ) = Θ(log(M f )), the two algorithms have the
same asymptotic time complexity. On the other hand, Fast-Rand-Elect requires
a user to fix the success probability (given by 1 − 1/f ) a priori.

6 Conclusion

In this paper, we have presented three leader election algorithms for multi chan-
nel radio networks with no collision detection. The deterministic algorithm guar-
antees leader election for all relevant channels (available to one or more nodes in
the network) and takes at most 2N +M log M  time slots, which is a considerable
improvement over the trivial algorithm which takes N M time slots for comple-
tion. The two randomized algorithms guarantee leader election for all relevant
channels with probability at least 1 − f1 within O(M log2 (nmax ) + M log2 (M f ))
and O(M log(nmax ) log(M f )) time slots, respectively.
Note that, when nmax = Θ(M ), the deterministic algorithm has time-
complexity of O(M log M ), which is asymptotically better than that of the ran-
domized algorithms. Moreover, with a randomized algorithm, there is always a
non-zero probability that a leader may not be elected for one or more of the rel-
evant channels. However, the deterministic algorithm guarantees that a leader
is elected for all the relevant channels with 100% confidence.
 Leader Election Algorithms for Multi-channel Wireless Networks 321

References
1. Akyildiz, I.F., Lee, W.Y., Vuran, M.C., Mohanty, S.: Next generation/dynamic
spectrum access/cognitive radio wireless networks: A survey. Computer Net-
works 50(13), 2127–2159 (2006)
2. Mitola, J.: Cognitive radio: An Integrated Agent Architecture for Software Defined
Radio. PhD thesis, Royal Institute of Technology (2000)
3. Perkins, C.E., Belding-Royer, E.M.: Ad-hoc on-demand distance vector routing. In:
Proceedings of the 2nd Workshop on Mobile Computing Systems and Applications
(WMCSA), pp. 90–100. IEEE Computer Society, Los Alamitos (1999)
4. DeCleene, B., Dondeti, L., Griffin, S., Hardjono, T., Kiwior, D., Kurose, J., Towsley,
D., Vasudevan, S., Zhang, C.: Secure group communication for wireless networks.
In: Proceedings of the IEEE Military Communications Conference (MILCOM), pp.
113–117 (2001)
5. Heinzelman, W.R., Chandrakasan, A., Balakrishnan, H.: Energy-efficient commu-
nication protocol for wireless microsensor networks. In: Proceedings of the Annual
Hawaii International Conference on System Sciences (HICSS) (2000)
6. Mittal, N., Krishnamurthy, S., Chandrasekaran, R., Venkatesan, S.: A Fast Deter-
ministic Algorithm for Neighbor Discovery in Multi-Channel Cognitive Radio Net-
works. Technical Report UTDCS-14-07, The University of Texas at Dallas (2007)
7. Krishnamurthy, S., Thoppian, M.R., Kuppa, S., Chandrasekaran, R., Mittal, N.,
Venkatesan, S., Prakash, R.: Time-efficient distributed layer-2 auto-configuration
for cognitive radio networks. Computer Networks 52(4), 831–849 (2008)
8. Bansal, T., Mittal, N., Venkatesan, S.: Leader Election Algorithms for Multi-
Channel Wireless Networks. Technical Report UTDCS-19-08, The University of
Texas at Dallas (2008)
9. Jurdzinski, T., Kutylowski, M., Zatopianski, J.: Weak communication in single-
hop radio networks: Adjusting algorithms to industrial standards. Concurrency
and Computation: Practice and Experience 15(11-12), 1117–1131 (2003)
10. Jurdzinski, T., Kutylowski, M., Zatopianski, J.: Efficient algorithms for leader elec-
tion in radio networks. In: Proceedings of the 21st ACM Symposium on Principles
of Distributed Computing (PODC), pp. 51–57 (2002)
11. Nakano, K., Olariu, S.: Randomized leader election protocols in radio networks
with no collision detection. In: Lee, D.T., Teng, S.-H. (eds.) ISAAC 2000. LNCS,
vol. 1969, pp. 362–373. Springer, Heidelberg (2000)
12. Nakano, K., Olariu, S.: Uniform leader election protocols for radio networks. IEEE
Trans. Parallel Distrib. Syst. 13(5), 516–526 (2002)
13. Bordim, J.L., Ito, Y., Nakano, K.: An energy efficient leader election protocol for
radio network with a single transceiver. IEICE Transactions 89-A(5), 1355–1361
(2006)
14. Nakano, K., Olariu, S.: A survey on leader election protocols for radio networks. In:
Proceedings of the International Symposium on Parallel Architectures, Algorithms
and Networks (ISPAN), pp. 63–68. IEEE Computer Society, Los Alamitos (2002)
15. Metcalfe, R., Boggs, D.: Ethernet: Distributed packet switching for local computer
networks. Commun. ACM 19(7), 395–404 (1976)
16. Hayashi, T., Nakano, K., Olariu, S.: Randomized initialization protocols for packet
radio networks. In: Proceedings of the 13th International Parallel and Processing
Symposium (IPPS), pp. 544–548. IEEE Computer Society, Los Alamitos (1999)